Library Note Pattern
Every once in a while I need to write a comment in one place and then refer to that comment from other places.
Turns out there is now a language-independent tool that helps
to use this pattern -
tagref!
tagref claims to be inspired by the Glasgow Haskell Compiler (GHC),
and specifically by their “Notes” system
described in
The Architecture of Open Source Applications
specifically chapter 5 section 6.
GHC Style Guide (a) also describes this pattern. Version of the style guide as of 2025-07-18 has exactly the same text as the book mentioned above. See archived link from 2020 for a slightly different phrasing.
Transcribing talks
I wanted to get an LLM summary for XOXO Festival. Lisa Hanawalt, BoJack Horseman - XOXO Festival (2015) which turns out to be a perfect opportunity to test out transcribing audio using Whisper!
Here is what I did to get a human readable transcript:
- Download the video
yt-dlp --extract-audio --audio-format mp3 'https://www.youtube.com/watch?v=f6F_CF7Yvo0' -o talk-audio.mp3
- Chunk into 10 minute chunks1
ffmpeg -i talk-audio.mp3 -f segment -segment_time 600 -c copy talk-audio-10m_chunks-%03d.mp3
- Transcribe the audio using Whisper through huggingface UI
- Clickity click in https://huggingface.co/openai/whisper-large-v3 ,
copy-pasting results into
transcript.txt.
- Clickity click in https://huggingface.co/openai/whisper-large-v3 ,
copy-pasting results into
- Chunk the text using Claude 3.5 Sonnet
cat transcript.txt | llm -m claude-3.5-sonnet -s "Split the content of this transcript up into paragraphs with logical breaks. Add newlines between each paragraph." > transcript-chunked.txt
This is already very useful!
Learning geometry through formal proofs
Every once in a while I try to understand something in detail, and often understanding something in detail requires understanding a bunch of math. I then go and try to understand the relevant math, but that depends on yet more math I don’t understand and so I give up.
The most recent iteration of this attempt at understanding something, started in December 2022, is my attempt at understanding “How Archimedes showed that π is approximately equal to 22/7” 1 .
Reading, notes, spaced repetition
Over the last year, I’ve stumbled on a series of texts that together changed how I approach learning.
“How to take smart notes” 1 introduced me to the idea that reading a text is insufficient to understand the ideas in the text. The suggested solution is, in short: write about what you read. Slightly more specifically: create a densely connected network of notes based on what you read.
“Why books don’t work” 2 re-iterates the “reading is insufficient” point, but also suggests that one needs more than to write, one must also remember. This text suggests using Spaced Repetition as a learning mechanism: you read, you write cards, you memorize cards using Spaced Repetition Software.